xen-devel.lists.xenproject.org archive mirror
 help / color / mirror / Atom feed
From: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
To: Bob Liu <bob.liu@oracle.com>
Cc: xen-devel@lists.xenproject.org, linux-kernel@vger.kernel.org,
	roger.pau@citrix.com
Subject: Re: [PATCH 1/2] xen-blkfront: don't call talk_to_blkback when already connected to blkback
Date: Tue, 7 Jun 2016 11:25:35 -0400	[thread overview]
Message-ID: <20160607152524.GA10281__26486.0852541917$1465313247$gmane$org@localhost.localdomain> (raw)
In-Reply-To: <574E7763.9060001@oracle.com>

On Wed, Jun 01, 2016 at 01:49:23PM +0800, Bob Liu wrote:
> 
> On 06/01/2016 04:33 AM, Konrad Rzeszutek Wilk wrote:
> > On Tue, May 31, 2016 at 04:59:16PM +0800, Bob Liu wrote:
> >> Sometimes blkfont may receive twice blkback_changed() notification after
> >> migration, then talk_to_blkback() will be called twice too and confused
> >> xen-blkback.
> > 
> > Could you enlighten the patch description by having some form of
> > state transition here? I am curious how you got the frontend
> > to get in XenbusStateConnected (via blkif_recover right) and then
> > the backend triggering the update once more?
> > 
> > Or is just a simple race - the backend moves from XenbusStateConnected->
> > XenbusStateConnected - which retriggers the frontend to hit in
> > blkback_changed the XenbusStateConnected state and go in there?
> > (That would be in conenct_ring changing the state). But I don't
> > see how the frontend_changed code get there as we have:
> > 
> >  770                 /*
> >  771                  * Ensure we connect even when two watches fire in
> >  772                  * close succession and we miss the intermediate value
> >  773                  * of frontend_state.
> >  774                  */
> >  775                 if (dev->state == XenbusStateConnected)
> >  776                         break;
> >  777 
> > 
> > ?
> > 
> > Now what about 'blkfront_connect' being called on the second time?
> > 
> > Ah, info->connected is probably by then in BLKIF_STATE_CONNECTED
> > (as blkif_recover changed) and we just reread the size of the disk.
> > 
> > Is that how about the flow goes?
> 
> 	blkfront 					blkback
> blkfront_resume()                                   
>  > talk_to_blkback()
>   > Set blkfront to XenbusStateInitialised
> 						Front changed()
>   						 > Connect()
> 						  > Set blkback to XenbusStateConnected
> 
> blkback_changed()
>  > Skip talk_to_blkback()
>    because frontstate == XenbusStateInitialised
>  > blkfront_connect()
>   > Set blkfront to XenbusStateConnected
> 
> 
> ------------------------------------------------------------------
> But sometimes blkfront receives
> blkback_changed() event more than once!

I think I know why. The udev scripts that get invoked when when
we attach a disk are a bit custom. As such I think they just
revalidate the size leading to this.

And this 'poke-at-XenbusStateConnected' state multiple times
is allowed. It is used to signal disk changes (or just to revalidate).
Hence it does not matter why really - we need to deal with this.

I modified your patch a bit and are testing it:

From e49dc9fc65eda4923b41d903ac51a7ddee182bcd Mon Sep 17 00:00:00 2001
From: Bob Liu <bob.liu@oracle.com>
Date: Tue, 7 Jun 2016 10:43:15 -0400
Subject: [PATCH] xen-blkfront: don't call talk_to_blkback when already
 connected to blkback

Sometimes blkfront may twice receive blkback_changed() notification
(XenbusStateConnected) after migration, which will cause
talk_to_blkback() to be called twice too and confuse xen-blkback.

The flow is as follow:
   blkfront                                        blkback
blkfront_resume()
 > talk_to_blkback()
  > Set blkfront to XenbusStateInitialised
                                                front changed()
                                                 > Connect()
                                                  > Set blkback to XenbusStateConnected

blkback_changed()
 > Skip talk_to_blkback()
   because frontstate == XenbusStateInitialised
 > blkfront_connect()
  > Set blkfront to XenbusStateConnected

-----
And here we get another XenbusStateConnected notification leading
to:
-----
blkback_changed()
 > because now frontstate != XenbusStateInitialised
   talk_to_blkback() is also called again
  > blkfront state changed from
  XenbusStateConnected to XenbusStateInitialised
    (Which is not correct!)

						front_changed():
                                                 > Do nothing because blkback
                                                   already in XenbusStateConnected

Now blkback is in XenbusStateConnected but blkfront is still
in XenbusStateInitialised - leading to no disks.

Poking of the XenbusStateConnected state is allowed (to deal with
block disk change) and has to be dealt with. The most likely
cause of this bug are custom udev scripts hooking up the disks
and then validating the size.

Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
---
 drivers/block/xen-blkfront.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index b4b8fbd..7765ad5 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -2484,10 +2484,23 @@ static void blkback_changed(struct xenbus_device *dev,
 		break;
 
 	case XenbusStateConnected:
-		if (dev->state != XenbusStateInitialised) {
+		/*
+		 * talk_to_blkback sets state to XenbusStateInitialised
+		 * and blkfront_connect sets it to XenbusStateConnected
+		 * (if connection went OK).
+		 *
+		 * If the backend (or toolstack) decides to poke at backend
+		 * state (and re-trigger the watch by setting the state repeatedly
+		 * to XenbusStateConnected (4)) we need to deal with this.
+		 * This is allowed as this is used to communicate to the guest
+		 * that the size of disk has changed!
+		 */
+		if ((dev->state != XenbusStateInitialised) &&
+		    (dev->state != XenbusStateConnected)) {
 			if (talk_to_blkback(dev, info))
 				break;
 		}
+
 		blkfront_connect(info);
 		break;
 
-- 
2.4.11


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

  parent reply	other threads:[~2016-06-07 15:25 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <1464685157-30738-1-git-send-email-bob.liu@oracle.com>
2016-05-31  8:59 ` [PATCH 2/2] xen-blkfront: fix resume issues Bob Liu
2016-05-31 20:33 ` [PATCH 1/2] xen-blkfront: don't call talk_to_blkback when already connected to blkback Konrad Rzeszutek Wilk
     [not found] ` <20160531203307.GC23808@char.us.oracle.com>
2016-06-01  5:49   ` Bob Liu
     [not found]   ` <574E7763.9060001@oracle.com>
2016-06-02 14:30     ` Konrad Rzeszutek Wilk
2016-06-07 15:25     ` Konrad Rzeszutek Wilk [this message]
     [not found]     ` <20160607152524.GA10281@localhost.localdomain>
2016-06-08  6:46       ` Bob Liu
     [not found]       ` <5757BF4E.9080307@oracle.com>
2016-06-08 14:47         ` Konrad Rzeszutek Wilk
     [not found]         ` <20160608144739.GD17027@char.us.oracle.com>
2016-06-15  8:39           ` Ross Lagerwall
     [not found]           ` <57611424.2020802@citrix.com>
2016-06-15 14:08             ` Konrad Rzeszutek Wilk
2016-05-31  8:59 Bob Liu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='20160607152524.GA10281__26486.0852541917$1465313247$gmane$org@localhost.localdomain' \
    --to=konrad.wilk@oracle.com \
    --cc=bob.liu@oracle.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=roger.pau@citrix.com \
    --cc=xen-devel@lists.xenproject.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).